Page 203 - The-5th-MCAIT2021-eProceeding
P. 203

computed in WordNet. The final semantic relatedness score l of two words that combines WordNet and word
        embedding is defined as in Eq.1:


                                                         1
            rel(w , w ) = max cos (v(w ), v(w )) + (1 − λ)                                                                 (1)
                 i
                   j
                                        j
                                   i
                       m,n λ                        dist(Sw , Sw )
                                                              j
                                                          i
        Where  dist  (Si,  Sj)  is  the  distance  between  two  senses  Si,  m,  and  Sj,  n.  v(w ), v(w )  are  the  vector
                                                                                   j
                                                                             i
        representations of word wi and wj in the word embedding. λ is a weighting factor. The final combined word to
        combined sense representation model is defined as in Eq. 2:

           W2CS(w) = {w |w ∶   rel(w , w ) > t }                                                                                                           (2)
                           j
                        j
                                   i
                                      j
                                           1

        3.2. Step 2: SRT enrichment to build Social Requirement Terms Vocabulary SRTV

           In this work, SynSet WordNet is used to enrich SRT by adding the synonymous words onto the existing
        keywords to achieve the Social Requirement Terms Vocabulary (SRTV). Eventually, it will increase the
        distinct gap between related reviews and non-related reviews in classifying the reviews of the applications.
        SRTV contains a group of words labeled to each term of SRT, as shown in Figure 2.
                       SRT                                SRTV                label
                        Privacy     Word                                       1
                                  Embedding       Privacy, confidentiality, ········
                                     &
                      Freedom       SynSet        Freedom, independence, ·······    2
                                    WordNet

                           ·
         Fig. 2. Enriching SRT to build SRTV.

        3.3. Step 3: Build BOR by relating SRTV label with SMA review
           The Bag-of-Requirements or BOR (as shown in Figure 3) is a SMA review that has been vectorized into
        BOW and n-gram and labeled with SRTV.  The semi-supervised KNN approach applied to the vectorized SRTV
        and reviews to achieve multi BOR labeling for each review.


                     Apps label                  Reviews                    SRTV label

                     Facebook     It forces me to put all my personal and private information   1,2,6
                                  including my phone number.

                     WhatsApp     It used to be a place to go to connect with friends and   2,4,9
                                  colleagues and there were freedoms used every day. Now it is
                                  not a place that you can speak freely.

          Fig. 3. Bag-of-Requirement data organization.






        E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021)   [190]
        Artificial Intelligence in the 4th Industrial Revolution
   198   199   200   201   202   203   204   205   206   207   208